Building ML systems using AWS

Before we can explore the fundamentals of how to implement the distributed training strategies highlighted at the outset, we first need to level set and understand just how the ML model training exercise can be performed on the AWS platform. Once we understand how AWS handles model training, we can further expand on this concept to address the concept of distributed training.

To assist ML practitioners in building ML systems, AWS provides the SageMaker (https://aws.amazon.com/sagemaker/) service. While SageMaker is a single AWS service, it comprises multiple modules that map specifically to an ML task. For example, SageMaker provides the Training job component that is purpose-built to take care of the heavy lifting and scaling of the model training task. ML practitioners can use SageMaker Training jobs to essentially provision ephemeral compute environments or clusters to handle the model training task. Essentially, all the ML practitioner needs to...